Block claims that sometimes we see more than we can report because the neural system for phenomenology “overflows” the system for accessibility. He makes the additional claim that this implies there are distinct neural mechanisms for phenomenology and cognitive accessibility. We argue that the difference in capacity between phenomenology and accessibility can be explained by noise amplification without any need to posit distinct systems. We explain why we think Block's approach is unable to build upon empirical findings, and suggest that metacognitive approaches will be more fruitful.

As a message passes down the line in the “broken telephone” or “Chinese whisper” game, it becomes garbled and some of its elements are completely lost. That is, the quality of information tends to deteriorate. Noise propagation and amplification also limit late sensory processing in the brain. This is why early forms of vision that are brief and iconic have larger capacities than later verbal reports which require deeper information processing. Simply put, the retina has more visual information for a simple visual perceptual event than the motor cortex. Hence, a difference in capacity is consistent with a single stream of serial processing and does not imply distinct processing systems, as Block claims. Note that our argument does not apply to situations when there are actually two streams of information, as in the dorsal-ventral distinction in visual processes (Goodale et al. Reference Goodale, Milner, Jakobson and Carey1991). In that case, the two systems are largely independent of each other. In Block's case though, the later cognitive access certainly depends on and receives its major inputs from the earlier brief processing, and for this reason the capacity difference is trivial.

Although we disagree with Block over the explanation of the differing capacities of phenomenology and cognitive accessibility, we agree that forced-choice reports can fail to capture what feels to be seen, especially when there is a lot going on in the visual presentation. This leads Block to propose “a neural mechanism by which phenomenology can overflow cognitive accessibility” (sect. 14, para. 3). But “overflow” is just one example of the failure of forced-choice reports. There are also cases in which forced-choice reports capture more than what is consciously seen. People with V1 lesions claim not to see anything in their affected visual field and yet make accurate visual discriminations; that is, they have blindsight (Weiskrantz Reference Weiskrantz1986). There are other cases in which forced-choice reports made with different modalities (e.g., manual button press, eye blinks, verbal reports) yield inconsistent measures of phenomenology given the same stimulus (Marcel Reference Marcel, Bock and Marsh1993). There are yet other cases in which a stimulus can cause people to make a forced-choice response that they do not want to make (Debner & Jacoby Reference Debner and Jacoby1994; Persaud & McLeod Reference Persaud and McLeod2007). So forced-choice reports are not ideal for measuring phenomenology. But this does not mean that we must associate phenomenology with a neural system that has a different processing capacity than that reflected by normal forced-choice reports. Nor does it mean that the capacity reflected by forced-choice reports under optimal cueing conditions (as in Sperling-style experiments) is the capacity for phenomenology. We never know, because forced-choice reports sometimes capture too much, sometimes capture too little, sometimes are inconsistent, and sometimes capture irrelevant information. We suggest that we must explore alternative measures, as it is vital to find reliable and valid ways of measuring phenomenology behaviourally before attempting to map it to a specific brain mechanism. We have been doing just this by employing metacognitive measures (Lau & Passingham Reference Lau and Passingham2006; Persaud et al. Reference Persaud, McLeod and Cowey2007); that is, we collect subjective reports, or judgements of performance, in addition to forced-choice reports regarding the stimuli.

Part of our motivation for using metacognitive measures is demonstrated by how Block's argument fails to find empirical support where he claims it does. Block claims that recurrent processing (feedback loops; for motion that is V1→V5→V1) within the visual cortices may support phenomenology. Given the above argument about capacities, it is clear that any processing stage prior to the stage that supports normal reportability would have a capacity large enough to “overflow” cognitive accesibililty, and thus be a good candidate for the supposed phenomenology. The retina, for instance, has all the visual information needed to support what is likely to be seen but not reported. Of course, Block does not think that the retina is a candidate. Presumably the reason is that the retina is not necessary for phenomenal vision: Electrical stimulation of the primary visual cortex can cause visual phenomenology without the retinal involvement. So being necessary for phenomenology is an important criterion. But pure feedforward processing (i.e., V1→V5, without the feedback for the case of motion) may fit this criterion as well, and, if one follows the above argument about capacities and inheritance of noise, the information capacity of this processing would certainly overflow cognitive accessibility.

Block attempts to support his feedback hypothesis by pointing out that disrupting feedback processing is correlated with a lack of visual consciousness (Pascual-Leone & Walsh Reference Pascual-Leone and Walsh2001). But if Block's argument that there can be stimuli which a person can see but not report is right, how do we know that in these cases there is a lack of phenomenology (and not just a lack of cognitive accessibility)? Block's argument backfires: If we allow for phenomenology without access, we would not be able to know when people do not see visual stimuli. For example, when magnetic fields disrupt feedback processing and people report not seeing stimuli they would otherwise see, how can we know that people do not actually see the stimuli? How can we know that feedback processing within the visual cortices does not just reflect cognitive access?

We believe that using alternative measures of phenomenology, such as metacognitive measures, may fill this gap. Although they may be imperfect, metacognitive measures are the best available method for determining when a person is aware of a stimulus. It is only after awareness can be properly measured that the neural substrates of consciousness can be found. Thus, metacognitive measures avoid the circularity inherent in Block's approach – that is, the very circularity in the “methodological puzzle of consciousness research” that Block attempts to address.